peace, terrorism, or civil conflict? understanding the decision of …ftp.iza.org/dp9996.pdf ·...
TRANSCRIPT
Forschungsinstitut zur Zukunft der ArbeitInstitute for the Study of Labor
DI
SC
US
SI
ON
P
AP
ER
S
ER
IE
S
Peace, Terrorism, or Civil Conflict?Understanding the Decision of an Opposition Group
IZA DP No. 9996
June 2016
Michael Jetter
Peace, Terrorism, or Civil Conflict? Understanding the Decision of an
Opposition Group
Michael Jetter University of Western Australia
and IZA
Discussion Paper No. 9996 June 2016
IZA
P.O. Box 7240 53072 Bonn
Germany
Phone: +49-228-3894-0 Fax: +49-228-3894-180
E-mail: [email protected]
Any opinions expressed here are those of the author(s) and not those of IZA. Research published in this series may include views on policy, but the institute itself takes no institutional policy positions. The IZA research network is committed to the IZA Guiding Principles of Research Integrity. The Institute for the Study of Labor (IZA) in Bonn is a local and virtual international research center and a place of communication between science, politics and business. IZA is an independent nonprofit organization supported by Deutsche Post Foundation. The center is associated with the University of Bonn and offers a stimulating research environment through its international network, workshops and conferences, data service, project support, research visits and doctoral program. IZA engages in (i) original and internationally competitive research in all fields of labor economics, (ii) development of policy concepts, and (iii) dissemination of research results and concepts to the interested public. IZA Discussion Papers often represent preliminary work and are circulated to encourage discussion. Citation of such a paper should account for its provisional character. A revised version may be available directly from the author.
IZA Discussion Paper No. 9996 June 2016
ABSTRACT
Peace, Terrorism, or Civil Conflict? Understanding the Decision of an Opposition Group*
When do opposition groups decide to mount a terrorism campaign and when do they enter an open civil conflict against the ruling government? This paper models an opposition group’s choice between peace, terrorism, and open conflict. Terrorism emerges if executive constraints are intermediate and rents are sizeable. Open conflict is predicted to emerge under poor executive constraints. Analyzing country-level panel data firmly supports these hypotheses, even when relying on within-country variation only in a fixed-effects framework. In particular, both the incidence of terrorism and the likelihood of terrorism onset increase under intermediate executive constraints (following an inverted U-shape) and if large rents are available from natural resources, oil, or foreign development assistance. A one-standard-deviation increase in rents raises casualties by approximately 15 percentage points. Related to civil conflict, moving from an authoritarian regime to comprehensive executive constraints is associated with a decrease in the number of battle-related deaths by approximately 74 percentage points. These findings can help us to better understand and anticipate the underlying decision of opposition groups and their choice between peace, a terrorism campaign, and open conflict. JEL Classification: D74, F35, O11, P47, Q34 Keywords: conflict, executive constraints, foreign aid, natural resource rents, oil rents,
political institutions, rents, terrorism Corresponding author: Michael Jetter Business School University of Western Australia 35 Stirling Highway Crawley, WA 6009 Australia E-mail: [email protected]
* I am grateful to Julia Debski for outstanding research assistance in this project.
1 Introduction
We continue to observe large-scale organized violence against ruling governments in many coun-
tries around the world. Sometimes, discontent materializes in an open insurgency against the
state, and other times domestic terrorism emerges. These phenomena are usually analyzed sep-
arately.1 In reality, however, an opposition group consciously decides about which (if any) form
of organized violence to pursue against a ruling government. Then why do some groups choose
an open insurgency against their government and others pursue a concealed strategy of domestic
terrorism?
Understanding the underlying drivers of large-scale organized violence against the state has
become more important than ever, as casualties are on the rise again. Figure 1 visualizes the
number of annual deaths from domestic conflicts and terrorism – both cracked record-highs in
2014 surpassing 100,000 and 42,000 casualties, respectively.
020
000
4000
060
000
8000
010
0000
Num
ber
of c
asua
lties
1970 1980 1990 2000 2010Year
Terrorism Conflicts
Figure 1: Worldwide casualties from terrorism and conflicts, using data from the UppsalaConflict Data Program (UCDP) and the Global Terrorism Database (GTD).
The following pages first introduce a basic theoretical intuition about how an opposition
group chooses its profit-maximizing strategy between (i) peace, (ii) terrorism, and (iii) open
insurgency. Although terrorism and conflicts share a number of common characteristics, the
1A notable exception is provided by Findley and Young (2012).
1
paper highlights an important distinction: whereas the potential gains and costs are large in open
insurgencies, both are naturally limited in terrorist campaigns. Two testable results emerge, as
the opposition’s choice depends on constraints on the executive and the available rents (e.g.,
natural resource rents, oil, or foreign development assistance), among other, mostly country-
specific parameters.
First, terrorism becomes likely if political constraints are intermediate and rents are sizeable.
Intuitively, the opposition does not want to risk losing its non-trivial share of rents in open
conflict, but the looming benefits from a terrorist campaign are more attractive than peace. In
reality, we have observed such scenarios in Algeria (Armed Islamic Group and Al-Qa’ida in the
lands of the Islamic Maghreb) over the past 20 years and in Nigeria (Boko Haram), for example.2
Interestingly, terrorism can emerge even under near-perfect institutions if rents are particularly
high.
Second, open conflict emerges if constraints on the executive are poor. In this case, the oppo-
sition has little to lose from mounting an open insurgency and rewards from a victorious uprising
are looming large. As examples, one may consider a number of domestic conflicts in Africa, such
as the Ethiopian civil war (1974 – 1991). Finally, peace prevails if (i) executive constraints
are sufficiently large and rents are moderate or (ii) executive constraints are well developed,
in which case the size of rents becomes irrelevant. Examples may be found in Scandinavian
nations, Australia, or New Zealand, among many others.
To test these hypotheses, I analyze country-year level data on casualties from terrorism
and domestic conflicts. Applying a two-way fixed-effects framework allows me to focus on
within-country variation only, controlling for any unobservable heterogeneity across countries
and time. Indeed, terrorism most likely occurs if a country exhibits an intermediate degree of
executive constraints and high rents, measured as natural resources (in particular oil) and foreign
development assistance. This quantitatively sizeable result holds for both the incidence and
onset of terrorism. In terms of magnitude, moving from an authoritarian regime to intermediate
political constraints translates to a 40 percentage point increase in casualties from terrorism.
2Algeria has maintained intermediate executive controls over the past 20 years with values ranging from threeto five on the Polity IV variable xconst (scale ranging from one to seven). The country also enjoys rents fromlarge oil and gas reserves, which it exports (see CIA, 2016). A similar scenario applies to Nigeria with executiveconstraints equivalent to a value of five since 1999 and sizeable oil exports.
2
Similarly, the likelihood of terrorism onset increases by approximately seven percentage points.
Consistent with the model, conflict intensity decreases linearly as institutional constraints on
the ruling elite improve. Moving from a completely authoritarian regime to inclusive institutions
is associated with a decrease in the number of deaths from domestic conflict by more than 74
percentage points. This relationship remains linear, as predicted by the model and confirms
previous results in the literature (see Hegre, 2014, for an overview).
The paper aims to contribute to two major streams in political science and political economy.
First, it provides a basic unifying framework to study terrorism and domestic conflict jointly.
This can help us understand (and potentially anticipate) why we observe terrorism in some
countries and open insurgency in others. It also explains why transitioning democracies can
become vulnerable to terrorism, namely if substantial rents are available. The simple theoretical
intuition builds on foundations from Besley and Persson (2009, 2011), who analyze political
violence from the government’s perspective, whereas this paper focuses on the choice of the
opposition group.
Second, the paper adds to empirical works on the determinants of terrorism and domestic
conflict. Terrorism can indeed be more likely in relatively democratic societies, as pointed
out by Chenoweth (2013), even after country-specific heterogeneity is accounted for. When
it comes to terrorism, intermediate constraints on the executive and sizeable rents can prove
a dangerous combination. Related to the civil conflict/civil war literature, the corresponding
findings highlight the overwhelming importance of institutional constraints, as previously argued
by Hegre (2014). The qualitative and quantitative interpretations of the derived findings are
considerable, both for the incidence and onset of terrorism and conflict.
The paper proceeds with a description of the literature, aimed at sorting this paper into
existing theories and empirical findings. Section 3 introduces a simple theoretical model of a
profit-maximizing opposition group, whereas section 4 presents the data and empirical method-
ology. Sections 5 and 6 analyze the empirical implications of the underlying hypotheses for
terrorism and civil conflict. Finally, section 7 concludes.
3
2 Background
Domestic terrorism and open insurgencies share many similarities as organized forms of political
violence. Blattman and Miguel (2010, p.6) point out that “the distinction between civil wars and
other forms of political instability has largely been assumed rather than demonstrated.” More
specifically, Lessing (2015) highlights the need to distinguish between forms of organized violence
that are intended to take control of the government, as opposed to those that carry other goals.
To clarify terminology, consider the respective definitions provided by the Merriam-Webster
dictionary:
• Terrorism: the use of violent acts to frighten the people in an area as a way of trying to
achieve a political goal.
• Insurgency: a usually violent attempt to take control of a government: a rebellion or
uprising.3
The uniting theme across these concepts centers on the notion of organized violence, usually
against a ruling government. At its roots, both concepts constitute expressions of a group’s
deep dissatisfaction with the status quo, leading them to choose violent means with the goal
of changing political institutions. However, a terrorist campaign and an open insurgency differ
along some relevant dimensions. Most importantly, terrorism is motivated by achieving a political
goal (generally speaking to obtain more political power), whereas the purpose of mounting an
insurgency lies in overthrowing the government completely.4
In economic terms, the costs and benefits differ between the concepts of terrorism and insur-
gency. The benefits from an open insurgency are larger (control of the government), but so are
the associated costs as an insurgency consists in open fighting against a ruling government. This
openness is expressed as a group’s public declaration of violent government opposition, which
in turn legitimizes the government’s persecution of all group members. As a consequence, the
group may lose its institutional privileges (little as they may have been) and become outlaws.
3Another term, virtually analogous to an insurgency relates to a political coup (or coup d’etat), defined as “asudden, violent, and illegal seizure of power from a government.”
4In practice, more political power may translate to regional independence, a more equal distribution of re-sources, or concessions in the country’s institutional framework.
4
In terrorism, on the other hand, resistance is usually organized underground, actors are
hidden, and the costs associated with terrorist attacks can be manageable, especially when com-
pared to mounting an insurgency. Usually, group members do not publicly identify with the
organization’s political goals and in the worst-case scenario those members conducting terrorist
missions will lose their institutionally guaranteed rights or die. The secret character of terror-
ism produces a natural, manageable limit to campaign costs. These distinctions between an
insurgency and terrorism will become important in section 3.
Previous works on the determinants of domestic conflicts usually distinguish between eco-
nomic (e.g., income levels) and political drivers (e.g., political rights or democracy).5 Some
studies have identified the presence of natural resources as a potential factor (e.g., see Collier
and Hoeffler, 1998, and Cotet and Tsui, 2013), as well as foreign aid inflows (e.g., see Nielsen
et al., 2011, and Nunn and Qian, 2014). The present article adds to these studies in highlighting
the importance of political constraints and rents from resources or international assistance in
explaining organized violence. In particular, the paper provides an explanation for the question
why some groups may choose terrorism over peace or open conflict.
Similar to conflict studies, the terrorism literature has identified some key drivers, such as de-
velopment levels (mostly GDP per capita) and political rights in several forms (e.g., democracy,
political freedom, civil liberties, or the rule of law).6 One particularly controversial observation
suggests that democratic states can become targets of terrorism – a phenomenon we do not
observe for domestic conflicts usually (see Hegre, 2014). Chenoweth (2013) writes that “tran-
sitioning democracies with internally inconsistent institutions were more likely to experience
5In a series of seminal papers, Paul Collier and Anke Hoeffler distinguish between greed and grievance, sug-gesting economic opportunity as the key driver of civil wars (Collier and Hoeffler, 1998, 2004; Collier et al., 2009).Miguel et al. (2004) show that higher income can alleviate conflict, using an instrumental variable approach basedon rainfall in Africa. Blattman and Miguel (2010) provide a comprehensive overview of the existing literature oncivil war. Dixon (2009) focuses on summarizing the empirical literature on the determinants of civil war onset.Further, the demographic composition of society, in particular ethnic fractionalization and polarization, has beenhighlighted in a number of influential papers, in particular by Fearon and Laitin (2003), Reynal-Querol and Mon-talvo (2005), and Joan Esteban and Debraj Ray (Esteban and Ray, 2008; Esteban and Ray, 2011; Esteban et al.,2012). In the present paper, ethnic components will not be the focus and will be assumed to remain constantwithin a country over time.
6The potential link between income levels and terrorism has received mixed evidence. Krueger and Maleckova(2003), Blomberg et al. (2004), Abadie (2006), and Enders and Hoover (2012) provide important studies. Notethat the present paper focuses on domestic terrorism, not international terrorism. In reality, only 3.8 percentof the documented terrorist attacks in the GTD are categorized as international terrorism. These missions areexcluded from the empirical analysis.
5
domestic terrorism than advanced democracies and authoritarian regimes.”
This paper provides an intuitive explanation for this observation. If constraints on the
executive are intermediate (as they are in countries moving from an authoritarian regime to
democracy) terrorism may emerge as the profit-maximizing choice of the opposition group. In
fact, even if institutional constraints are considerable, terrorism can be observed, but only if
rents are particularly large. These hypotheses emerge directly from the cost-benefit distinctions
between open conflict and concealed terrorism.
3 Modelling the Opposition’s Choice
Assume an opposition group, normalized to the size of one. To keep things simple, the size
of the opposition group equals the size of the ruling government group (akin to Besley and
Persson, 2009, 2011). Similarly, no within-group coordination problems are permitted, although
one could amend the decision process with such dynamics without loss of generality. Given the
status quo, which will be introduced shortly, the opposition can choose between one of three
strategies: peace, terrorism, or open insurgency. The decision process is modeled in its simplest
form as one static period for a risk-neutral opposition group, in order to emphasize the basic
underlying problem.
Following seminal models by Timothy Besley and Torsten Persson (2009 and 2011), the
country’s institutional foundations guarantee the opposition group a minimum share σ (with
0 ≤ σ ≤ 12) of the available rents, R. (It is straightforward to show that a ruling government
will always choose the lowest σ possible in the following framework. See the appendix for details.)
σ can be interpreted as constraints on the executive or, alternatively, as extractive (low σ) versus
inclusive institutions (high σ), following the terminology used by Acemoglu et al. (2005). In
the spirit of Besley and Persson (2009, 2011), R refers to natural resource rents or foreign aid
inflows – both assets over which a ruling government maintains control.
In reality, we observe governments that extract parts of these available rents in many coun-
tries. In the given setup, 1− σ represents the maximum share of R the ruling government can
extract. If σ = 12 , then rents are shared equally between both equally-sized groups and insti-
tutions are completely inclusive, i.e., the ruling elite does not extract any rents beyond their
6
proportional share. Contemporary examples are provided by a number of largely democratic
states, such as the majority of European states, North America, Australia, or New Zealand.
In the other extreme (σ = 0), the ruling group extracts all rents, leaving none for the
opposition. Examples can be found in strict authoritarian regimes, such as North Korea, Fidel
Castro’s Cuba, or Robert Mugabe’s Zimbabwe. In reality, most regimes fall somewhere in the
middle with 0 < σ < 12 .
3.1 Peace
The opposition group’s first option is characterized by non-violence. In particular, maintaining
peace will yield the opposition group a profit of
Πpeace = σR. (1)
Notice that if constraints on the executive are optimal and σ = 12 , then both groups benefit
equally from revenues, since the ruling group’s revenue remains (1−σ)R in this basic framework.
If σ = 0, however, no institutional restraints are posed on the executive and the reigning group
can reap all resource revenue, equivalent to R.
3.2 Terrorism
Now consider the second option: a terrorist campaign. The idea of forming a terrorist movement
consists in using organized violence to enforce better institutional terms, i.e., an even distribution
of the available rents, corresponding to σ = 12 . We can think of a number of potential demands
that can be summarized under σ, such as territorial concessions (e.g., separatist groups) or
improved political and economic power. However, it is not possible to take control of the
government with terrorist tactics, naturally limiting the potential gains of a terrorist campaign.
In turn, the costs of terrorism are determined by a fixed non-negative amount c for two
main reasons. First, mounting a terrorist campaign means that no fractional member of the
opposition needs to openly declare herself as violently opposed to the government. Terrorism
implies secrecy. Thus, accountability is limited by the concealed nature of terrorism. Second, in
reality, conducting terrorist attacks can be as simple as one individual firing a gun in a highly
7
populated area. This aspect of terrorism drastically limits the associated costs, especially when
compared to open insurgency.
Both of these characteristics capture the fundamental difference between terrorism and in-
surgencies. In general, the cost of terrorist attacks remains relatively small across countries and
time, although country- and time-specific aspects are likely to influence c.7 In the empirical
framework, fixed effects will be introduced to capture such unobservable heterogeneity across
countries and time. Further, if β (with 0 < β < 1) represents the probability of success, the
expected profit of mounting a terrorist campaign becomes8
Πterror = σR+ β(1
2− σ)R− c. (2)
In practice, β can include a number of country-specific aspects that may favor or complicate a
successful campaign of violence against the government. For instance, if the state’s institutions
are weak, the chances of a successful revolution increase (e.g., see discussion in Besley and
Persson, 2011). As another example, geographical aspects could facilitate or complicate the
chances of a successful revolution (e.g., see Abadie, 2006, for terrorism; Fearon and Laitin, 2003,
Collier et al., 2009, Do and Iyer, 2010, Weidmann and Ward, 2010, or Schutte, 2015, for conflict).
3.3 Insurgency
Finally, the opposition group can consider a third option: open insurgency. In this case, potential
gains are higher than from choosing a terrorist campaign, as the looming reward consists of taking
over the government and reaping (1 − σ)R. However, the costs of mounting an insurgency are
also non-trivial and amount to devoting all available resources into fighting (σR), as the group
declares an open war on the ruling regime. Another interpretation of the associated costs
amounting to σR relates to the notion that once a group declares war on the government it is
not eligible to receive its institutional share of rents anymore. Thus, the payoff from insurgency
becomes
7For example, we can think of technological advancements over time, country-specific government surveillance,or even within-group coordination problems as drivers of c.
8To be entirely accurate, this payoff function is an expected payoff function. However, since I am assuming arisk-neutral opposition group this distinction between expected and actual profit becomes irrelevant in the presentcontext.
8
Πwar = σR+ β(1− σ)R− σR. (3)
Note that this setup implies two simplifying assumptions. First, the probability of a suc-
cessful terrorist campaign is equalized to the probability of a successful insurgency. In reality,
of course, these odds may differ for several reasons and βterror = βwar = β is chosen for math-
ematical convenience. Second, an insurgency (and likely terrorism as well) may destroy rents,
potentially reducing R. Nevertheless, altering either assumption would not change the derived
qualitative implications, but would complicate calculations. The general intuition of this mod-
eling is preserved for βterror 6= βwar and for introducing an additional cost term measuring
destruction from war. In the empirical section, country- and time-fixed effects are employed to
control for such heterogeneity within countries and time periods.
3.4 Optimal Choice of the Opposition
Which option will the opposition group choose? First, it is useful to review corner solutions,
i.e., cases where institutional constraints are completely absent or ideal. In reality, it is highly
unlikely that the group will choose insurgency if σ → 12 , as we rarely observe attempts aimed at
overthrowing the government in societies with inclusive institutions (see Hegre, 2014). Indeed,
β < 1 is sufficient to ensure Πpeace > Πwar for σ = 12 .
In addition, it is realistic to assume that at least one violent option dominates the peaceful
scenario if σ = 0. Note that equation 3 by definition fulfills that restriction, whereas
R >2c
β(4)
would be required for terrorism to even be considered. Intuitively, if available rents are suf-
ficiently small, terrorism does not present itself as a lucrative option at all. Similarly, if the
associated costs (c) of mounting a terrorist campaign are large, terrorism becomes a less desir-
able alternative. Finally, if the probability of success (β) approaches zero, terrorism loses its
attractiveness for the opposition.9
9In fact, the idea of not negotiating with terrorists aims to reduce βterror to zero, signaling that terrorist actswould by no means be considered in the political process. However, governments have frequently broken thatideal. For more detailed analyses, see Clutterbuck (1992), Neumann (2007), and Dolnik and Fitzgerald (2011).
9
Comparing payoffs, we can then derive the conditions under which each option is preferable.
Figure 2 visualizes the potential scenarios, graphing the constraints on the executive (σ) to the
available rents (R). In particular, we can establish the following optimal strategies of violence:
Proposition 1. If executive constraints are intermediate (with cR + 1
2β < σ < 12 −
cβR)
and rents are non-trivial (with R > 2c(1+β)β(1−β) ), terrorism becomes an attractive option
for the opposition group.
Proof. Follows directly from comparing equation 2 to equations 1 and 3.
Figure 2: Predicted occurrence of war, terrorism, and peace under different combinations ofexecutive constraints and available rents.
Comparing terrorism to peace, even if σ approaches 12 an opposition may find it beneficial to
promote terrorism, but only if rents are particularly large. This may explain why even relatively
10
advanced democracies can suffer from domestic terrorism. Comparing terrorism to an insurgency
suggests that if constraints are exceptionally poor, then the opposition has little to lose from
an open conflict. On the other hand, if σ is moderate already, then seeking a larger share of R
via terrorist measures can prove to be more beneficial, as the potential losses from an open war
loom too large.
Proposition 2. If executive constraints are small (with σ < β1+β and σ < c
R + 12β)
insurgency becomes likely.
Intuitively, the decision between peace and insurgency is simple: if the share of rents received
is exceptionally low, then an opposition group has little to lose from mounting an insurgency.
This becomes more likely if the state’s power is weak (implying a high β). This explains why
countries with weak institutions struggle to maintain peace – the chances for a successful rev-
olution (β) are just too attractive and usually executive constraints are too weak (i.e., a small
σ). These hypotheses have received firm support in the empirical literature (e.g., see Hegre and
Sambanis, 2006, and Blattman and Miguel, 2010, for comprehensive summaries).
4 Data and Empirical Methodology
4.1 Data
The literature generally considers three aspects of violent conflicts: incidence, onset, and du-
ration (see Blattman and Miguel, 2010, for a detailed discussion). Naturally, the provided
theoretical intuition is most applicable to the incidence and onset of terrorism and conflict, as
studies on the duration of such events are likely to follow different dynamics (e.g., see Acemoglu
et al., 2010, and Acemoglu and Wolitzky, 2014, for recent theoretical works). Thus, the empirical
section will focus on studying incidence and onset of terrorism, followed by the same sequence
for civil conflict.
To test Propositions 1 and 2, I access the GTD (introduced by LaFree and Dugan, 2007) and
the UCDP (UCDP, 2015) for detailed data on deaths from terrorist attacks and internal conflicts.
The GTD (START, 2015) defines terrorism as “the threatened or actual use of illegal force and
violence by a non-state actor to attain a political, economic, religious, or social goal through fear,
11
coercion, or intimidation.” In particular, I will focus on the number of casualties from terrorism.
(Nevertheless, all derived results are consistent when using the number of attacks.) In turn,
the UCDP (UCDP, 2015) defines armed conflict as “a contested incompatibility that concerns
government and/or territory where the use of armed force between two parties, of which at least
one is the government of a state, results in at least 25 battle-related deaths.”
The GTD contains information on terrorist attacks from 1970 to 2014 and I aggregate that
information to the country-year level.10 The UCDP battle-related deaths dataset provides infor-
mation on the number of casualties from internal and internationalized internal conflicts on the
country-year level from 1989 to 2014. Both data sources have become standard in the respective
literature. Table AII lists all sample countries with their respective number of observations for
both samples.
Note that the literature generally refers to conflicts with more than 1,000 battle-related
deaths in a given year as civil wars (e.g., see Blattman and Miguel, 2010). If a threshold of 25
battle-related deaths has been crossed, then researchers refer to civil conflict. Most datasets only
offer binary indicators for civil conflict and civil war. However, the UCDP battle-related deaths
dataset allows for a much more continuous measure of an open and violent opposition to the
government, providing the number of casualties. Nevertheless, all results are consistent when
using a binary indicator for more than 25 casualties as the dependent variable for terrorism and
conflict. The corresponding findings are referred to Table AV in the appendix.
To measure σ, the institutional constraints on the executive, I access the Polity IV dataset, a
common source for political variables (introduced by Marshall and Jaggers, 2002). In particular,
I use the variable xconst (executive constraints, labeled EXEC from hereon), ranging from one
to seven, where larger values symbolize tighter constraints on the executive. In alternative
estimations, I also employ the polity2 variable capturing a country’s degree of democracy.
Further, measures forR (natural resource rents, foreign development assistance, and oil rents)
are collected from the World Development Indicators (Group, 2012). Finally, additional control
variables are taken from conventional sources for country-level data and will be introduced in the
upcoming subsection. Summary statistics of all variables are referred to the Appendix Tables
10The only missing year is 1993, in which the GTD does not provide data because of missing files. Note thatI exclude international terrorism and inter-country conflicts, as these phenomena are likely following differentdynamics than the domestic situation described in this paper.
12
AIII and AIV.
4.2 Empirical Methodology
4.2.1 The Incidence of Terrorism and Conflict
Analyzing the incidence of terrorism, I estimate the following regression for country i and year
t, before employing the same structure to estimate the incidence of conflicts.
Ln(1 + deaths) = α1EXECit + α2
(EXEC)2
it + α3Rentsit + Xitα4 + δi + ρt + κit + εit. (5)
Note that the dependent variable is calculated as Ln(1+deaths), which conserves country-year
observations where no deaths occurred. In the case of terrorism, we would expect α1 to exhibit
a positive coefficient, whereas α2 is predicted to be negative, corresponding to the notion that
terrorism is most likely to occur in societies with intermediate controls on the executive. For
conflicts, we would expect α1 to exhibit a negative coefficient and α2 should be statistically
irrelevant.
The effect of available rents is captured by α3 and in the case of terrorism we predict a
positive relationship. In particular, I will consider natural resource rents, oil rents, and foreign
development assistance as measures for the available rents (akin to Besley and Persson, 2009,
2011). Following the theoretical motivation, resource rents are expected to be less of a factor in
driving conflict.
To control for potentially confounding characteristics that may independently influence the
occurrence of large-scale organized violence against the state, the vector Xit incorporates the
conventional time-variant control variables. In particular, I include GDP per capita, population
size (employing the natural logarithm for both), and the rate of economic growth. These factors
have emerged as likely drivers of organized violence in the associated literature.11
δi and ρt constitute country- and year-fixed effects, whereas κit incorporates continent-
specific time trends. Note that country-fixed effects are absorbing any time-invariant country-
11For the importance of income and population size in explaining conflicts, see Collier and Hoeffler (1998),Fearon and Laitin (2003), or Cotet and Tsui (2013). Blomberg et al. (2004) and Enders and Hoover (2012)highlight the role of income levels in explaining terrorism. Blomberg et al. (2004) and Miguel et al. (2004) findgrowth rates to matter for terrorism and conflicts, respectively.
13
specific factors that could be associated with c and β. This captures geographical aspects,
colonial origin, individual history, and other time-invariant heterogeneity on the country level.
Fixed effects also reasonably control for characteristics that only change slowly over time, such
as ethnic shares or religious distributions. Thus, fixed effects are alleviating concerns about
omitted variables.
In addition, fixed effects provide a reasonable assurance against endogeneity concerns from
measurement error. For example, if data quality in certain (potentially less developed) countries
or specific timeframes is imprecise, a fixed-effects framework would capture such shortcomings.
In general, several topics of interest in the cross-country literature have encountered the im-
portance of using a fixed-effects framework to contain endogeneity concerns in a powerful way.
Examples can be found in the analysis of economic growth (see Islam, 1995) or democracy
(see Acemoglu et al., 2008, and Cervellati et al., 2014). In the case of understanding conflict
drivers, Cotet and Tsui (2013) have shown the importance of using panel data with fixed ef-
fects. Continent-specific time trends incorporate the idea that developments related to conflict
or terrorism can sometimes spill over into neighboring countries.12 The Arab Spring provides
a recent popular example. Finally, εit stands for the conventional error term, clustered on the
country level.
Beyond the incidence measures, the empirical analysis then turns to analyzing the onset of
terrorism and conflict. I first calculate a binary dependent variable that takes on the value of
one if a country suffers deaths from terrorism in a given year, but has not experienced such
deaths in the previous year. Measuring conflict onset follows the same logic.
Applying probit regressions allows me to estimate the influence of executive constraints and
rents on the likelihood of terrorism and conflict onset. As independent variables, I incorporate
the same regressors as in equation 5, excluding fixed effects and time trends. In practice, the
onset of both terrorism and conflict are (thankfully) much rarer within countries over time than
the variation in the incidence variable of casualties. As a consequence, a fixed-effects framework
would not leave sufficient statistical variation in the data to reveal the underlying relationships.
12In alternative estimations, I also incorporate country-specific time trends, producing a much tighter econo-metric framework. In these estimations, results are consistent with the displayed results in terms of suggestedsigns and magnitudes. In few estimations, the level of statistical significance decreases for some covariates. How-ever, this is to be expected, as time variation within a country is limited in a number of control variables. Thus,introducing country-specific time trends can absorb much of the underlying variation.
14
In fact, the average sample country incurs approximately 3.5 terrorism onset years and 0.64
conflict onset years. These aspects will be discussed in more detail as the corresponding results
are presented.
Finally, results from several alternative specifications will be presented for each estimation,
focusing on alternative measures for the key variables, as well as addressing potential endogeneity
concerns.
5 Empirical Analysis of Terrorism
This section discusses the empirical findings related to the incidence and onset of terrorism,
including alternative estimations. The same structure follows for the incidence and onset of
conflicts.
5.1 Incidence of Terrorism: Main Results
Table 1 considers the incidence of terrorism, measured by the number of casualties from terrorism
in country i and year t. In the first column, only a linear term of institutional constraints is
used as an explanatory variable. The derived coefficient is positive, but not relevant on any
conventional level of statistical significance. Were we to stop here, we would conclude that
executive constraints are unrelated to the incidence of terrorism.
Column (2) then acknowledges the nonlinearity suggested by the theoretical intuition. In-
deed, we find strong evidence for a quadratic shape and the respective coefficients are both
significant on the one percent level. As constraints on the executive strengthen, terrorism is
suggested to rise at first and then fall after peaking at a value of 4.3 on a scale of one to seven.
Column (3) includes country-fixed effects, thereby controlling for individual particularities of
each state. In the context of the basic model presented before, this controls for (but is not limited
to) the cost of conducting a terrorist attack (c) and the probability of a successful campaign
(β). It is interesting to see that the coefficients associated with institutional constraints only
change marginally, even though the explanatory power of the model increases substantially from
an adjusted R2 of 0.015 to explaining over 45 percent of the variation in the occurrence of
terrorism.
15
Table 1: OLS regression results, estimating the number of deaths from terrorism in country iand year t.
(1) (2) (3) (4) (5) (6)
Dependent variable: Ln(1+deaths from terrorism)
EXEC 0.017 0.508∗∗∗ 0.588∗∗∗ 0.372∗∗ 0.366∗∗ 0.308∗
(0.033) (0.166) (0.187) (0.178) (0.176) (0.164)
(EXEC)2 -0.059∗∗∗ -0.066∗∗∗ -0.047∗∗ -0.045∗∗ -0.036∗
(0.021) (0.023) (0.023) (0.022) (0.020)
Natural resource rents 0.485∗∗ 0.507∗∗ 0.669∗∗∗
in US$ 10,000/cap (0.236) (0.214) (0.233)
Country-fixed effects yes yes yes yes
Control variablesa yes yes yes
Year-fixed effects yes yes
Continent-specific time trends yes
# of countries 158 158 158 158 158 158N 5,400 5,400 5,400 5,400 5,400 5,400Adjusted R2 0.000 0.015 0.452 0.467 0.498 0.530
Notes: Standard errors clustered on the country level are displayed in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗
p < 0.01. aIncludes GDP/capita, population size, and growth rate.
16
Column (4) incorporates further control variables, in particular GDP per capita, population
size, the economic growth rate, and finally natural resource rents. Recall that natural resource
rents correspond to R in the model and we would expect a positive coefficient here. Indeed, this
hypothesis is supported.
Finally, adding year-fixed effects and continent-specific time trends provides a much tighter
econometric framework. In the most complete estimation, both underlying hypotheses are con-
firmed: constraints on the executive remain non-linear in predicting terrorism with a maximum
for intermediate ranges of institutional constraints, whereas higher natural resource rents are
associated with more terrorism.
In terms of magnitude, both results are non-trivial, as visualized in Figure 3. In partic-
ular, terrorism peaks at a value of 4.3 on the scale of executive constraints. Relative to a
completely authoritarian regime (value of one), the average number of deaths from terrorism is
approximately 40 percentage points higher in the case of intermediate institutional constraints.
Note that terrorism in largely inclusive institutions, corresponding to a value of σ close to 0.5,
still remains more prevalent than in authoritarian regimes. This finding is consistent with the
model’s predictions. Related to R, increasing natural resource rents by one standard deviation
(US$ 2,426 per capita) corresponds to a 16.2 percentage point increase in the casualties from
terrorism.
3040
5060
70%
cha
nge
in c
asua
lties
from
terr
oris
m
0 2 4 6 8Constraints on executive
020
4060
80%
cha
nge
in c
asua
lties
from
terr
oris
m
0 .2 .4 .6 .8 1Natural resource rents in $10,000 per capita
Figure 3: Effect of constraints on the executive and natural resource rents on the incidence ofterrorism, plotting results from column (6) of Table 1.
17
5.2 Incidence of Terrorism: Extensions
From these baseline findings related to the incidence of terrorism, I now consider several ex-
tensions, displayed in Table 2. Columns (1) and (2) turn to alternative measures for rents,
namely foreign development assistance and oil rents. In the spirit of Besley and Persson (2009,
2011), rents may relate to natural resources or foreign assistance – both of which are likely
at the disposal of a ruling government. First, including development assistance produces the
expected result, as larger inflows are associated with more terrorism. In terms of magnitude,
a one standard deviation increase in foreign assistance (US$ 117 per capita) corresponds to a
14.9 percentage point increase in the number of terrorism casualties. Throughout the additional
estimations in Table 2, this result remains remarkably stable.
Second, with respect to specific natural resources, column (2) supports the idea that larger
oil revenues directly correspond to more terrorism. (Note that to avoid multicollinearity issues,
I remove the measure for natural resource rents once oil rents are included. These variables
are highly correlated with a coefficient of 0.97.) In this case, a one standard deviation increase
(US$ 2,078) is associated with a 15.8 percentage point increase in the number of deaths from
terrorism.
Reminding ourselves of the previous coefficients associated with a one standard deviation
increase in natural resources (16.2 percentage points) or development assistance (14.9 percentage
points), these estimates are remarkably close. Thus, a general relationship between R and ter-
rorism appears likely, as suggested by Proposition 1. In addition, the non-linear result associated
with institutional constraints remains robust to these alternative estimations.
Focusing on the measure of institutional constraints, column (3) introduces an alternative
measure with the polity2 variable of democracy, provided by the Polity IV project. In order to
properly estimate the quadratic effect, I re-scale the initial polity2 variable to all positive values
ranging from zero (corresponding to total autocracy) to 20 (total democracy). It is interesting
to see that the derived result remains consistent and, if anything, statistical precision increases.
It is likely that a more detailed measure of institutional constraints, in which 20 degrees of
democracy are possible, contributes to a more precise estimation of the underlying relationship.
Note also that the corresponding results for development assistance and oil rents remain robust.13
13Incorporating natural resource rents instead of oil rents produces the same conclusion.
18
Table 2: OLS regression results from extensions, estimating the number of deaths from terror-ism in country i and year t.
IV regressionsb
(1) (2) (3) (4) (5) (6)
Dependent variable: Ln(1+deaths from terrorism)
EXEC 0.318∗ 0.298∗ 0.377∗ 0.355(0.172) (0.175) (0.212) (0.221)
(EXEC)2 -0.040∗ -0.038∗ -0.052∗ -0.051∗
(0.021) (0.022) (0.027) (0.028)
Natural resource rents 0.761∗∗∗ 0.522∗
in US$ 10,000/cap (0.279) (0.304)
Development assistance 12.701∗∗ 12.729∗∗ 10.450∗∗ 15.266∗ 16.049∗ 13.056∗∗
in US$ 10,000/cap (6.082) (6.146) (4.777) (8.879) (9.174) (6.404)
Oil rents 0.761∗∗ 0.619∗∗ 0.648∗∗ 0.355in US$ 10,000/cap (0.299) (0.252) (0.290) (0.290)
Polity IV 0.149∗∗ 0.247∗∗∗
(0.065) (0.086)
(Polity IV)2 -0.007∗∗ -0.012∗∗∗
(0.003) (0.004)
Control variablesa yes yes yes yes yes yes
Country-fixed effects yes yes yes yes yes yes
Year-fixed effects yes yes yes yes yes yes
Continent-specific time trends yes yes yes
# of countries 136 136 135 135 135 134N 4,333 4,190 4,157 4,229 4,049 4,014Adjusted R2 0.532 0.537 0.524 0.498 0.501 0.493
Notes: Standard errors clustered on the country level are displayed in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗
p < 0.01. aIncludes GDP/capita, population size, and growth rate. bIn column (4), EXEC, (EXEC)2, natural
resource rents, and development assistance are instrumented by their respective lagged values in the previous
year. Columns (5) and (6) apply the same logic for EXEC, (EXEC)2, oil rents, development assistance, Polity
IV and (Polity IV)2. In all estimations, Shea’s partial R2 ranges between 0.55 and 0.78, leaving little concern
about potentially weak instruments.
19
Finally, I conduct robustness checks using alternative measures for the dependent variable. In
particular, the benchmark results are consistent when employing a measure of deaths per capita
or estimating a more traditional binary indicator of 25 or more casualties. The corresponding
results are referred to the appendix Table AV.14
Finally, columns (4) to (6) display results from instrumental variable regressions, addressing
potential reverse causality concerns. For example, it is possible that pressure from terrorism in
turn affects the institutional equilibrium of a country, the degree of resource extraction, or the
level of international assistance. If that were the case, the coefficients derived in Table 1 could
be biased.15
To alleviate such concerns, I follow recent macroeconomic studies in using lagged values
of the variables of interest as instruments. In particular, the growth literature has resorted
to this technique (e.g., Temple, 1999; Schularick and Steger, 2010; Mirestean and Tsangarides,
2016), as well as studies analyzing effects from democracy (Bhattacharyya and Hodler, 2010) and
corruption (Arezki and Bruckner, 2011). This IV-approach, in combination with the theoretical
intuition provided in section 3 and the fixed-effects framework, should mitigate endogeneity
concerns.
Note that the derived coefficients in columns (4) to (6) are largely in line with the benchmark
OLS results from Table 1.16 In terms of statistical power, executive constraints turn marginally
insignificant on conventional levels (t-statistic of 1.61) in column (5), yet inflated standard errors
are likely to blame. In terms of magnitude, the coefficient associated with EXEC remains strong
and even rises (from 0.308 in column (6), Table 1, to 0.355).
Further, the importance of R prevails throughout the IV-estimations, with the exception of
oil rents in column (6). Nevertheless, a quantitative interpretation of the derived coefficient still
suggests a positive relationship between oil rents and terrorism. Finally, employing the polity2
14As regressions are estimated in a fixed-effects framework, I refrain from using a logit or probit approach toestimate the binary outcome variable of conflict incidence, but rather employ a conventional OLS approach, as iscommon in the literature (see Greene, 2004).
15Note that finding an instrumental variable for any given country on the yearly level provides a difficult task.Large country-specific shocks, such as colonialism or geography, are unsuitable candidates in a panel framework,as they provide no within-country variation. Other prominent candidates, such as natural disasters, have beenshown to directly affect income levels and conflict incidence, rendering them invalid for the present estimation.
16Testing for weak instruments confirms the validity of the lagged instruments, as Shea’s partial R2 statisticproduces values between 0.55 and 0.78 (see Shea, 1997).
20
variable produces a result that is consistent with the findings from OLS regressions. Overall,
these IV estimations confirm the findings derived in Table 1.
5.3 Onset of Terrorism
From terrorist incidence, I now move to probit regressions, predicting terrorist onset in Table
3 (displaying marginal effects). As before, a linear term is not sufficient to accurately describe
the underlying relationship, but column (2) produces the familiar nonlinearity once a quadratic
term is added. Note that the entire sample “only” produces 560 country-year observations in
which terrorism occurs, but has not occurred the year before. On average, this corresponds
to approximately 3.6 observations per country, indicating that incorporating fixed effects may
not leave sufficient statistical variation to reveal the underlying dynamics. (Nevertheless, a
fixed-effects framework produces the same quantitative conclusions, consistent with the results
displayed in Table 3.17)
The regression shown in column (3) includes the familiar control variables, in addition to
natural resource rents and development assistance. Consistent with the findings related to ter-
rorism incidence, development assistance emerges as a positive predictor. Although the measure
for natural resource rents barely misses the conventional hurdle of statistical significance (t-value
of 1.61), a sizeable positive coefficient is established. In addition, the familiar inverted U-shape
for the effect of institutional constraints on terrorism onset prevails.
Column (4) substitutes oil rents for overall natural resources, and we find the familiar positive
relationship to terrorism. Further, column (5) turns to the alternative measure for institutional
constraints by employing the polity2 variable. As before, the corresponding findings support all
predictions related to σ and R.
Finally, columns (6) to (8) display results from IV regressions, following the same sequence
as columns (4) to (6) in Table 2. It is reassuring to see that all suggested relationships re-
ceive strong support. Figure 4 plots the underlying relationships for executive constraints and
development assistance, using the results from column (3) as a reference point. Compared to
authoritarianism, terrorism becomes approximately seven percentage points more likely if con-
17When including country-fixed effects, EXEC produces a coefficient of 0.210 (standard error 0.114), whereas(EXEC)2 produces a coefficient of -0.022 (0.014). Natural resource rents produce a coefficient of 0.007 (0.004).
21
Tab
le3:
Res
ult
sfr
omp
rob
itre
gre
ssio
n,
esti
mat
ing
the
onse
tof
terr
oris
min
cou
ntr
yi
and
yeart.
Dis
pla
yin
gm
argi
nal
effec
ts.
IVb
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
Depen
den
tvariable:
Onsetofterrorism
EX
EC
0.0
04
0.0
61∗∗∗
0.0
42∗∗∗
0.0
40∗∗
0.2
14∗∗∗
0.2
03∗∗∗
(0.0
02)
(0.0
14)
(0.0
15)
(0.0
16)
(0.0
72)
(0.0
74)
(EX
EC
)2-0
.007∗∗∗
-0.0
04∗∗
-0.0
04∗∗
-0.0
21∗∗
-0.0
20∗∗
(0.0
02)
(0.0
02)
(0.0
02)
(0.0
09)
(0.0
09)
Natu
ral
reso
urc
ere
nts
0.0
61
0.3
05∗
0.3
10∗∗
inU
S$
10,0
00/ca
p(0
.038)
(0.1
75)
(0.1
31)
Dev
elopm
ent
ass
ista
nce
1.3
79∗∗
1.3
47∗∗
1.6
74∗∗∗
8.5
68∗∗∗
7.9
34∗∗∗
inU
S$
10,0
00/ca
p(0
.538)
(0.5
52)
(0.5
59)
(2.6
70)
(2.7
37)
Oil
rents
0.0
74∗
0.1
02∗∗∗
0.3
56∗
inU
S$
10,0
00/ca
p(0
.041)
(0.0
39)
(0.1
88)
Polity
IV0.0
31∗∗∗
0.1
48∗∗∗
(0.0
05)
(0.0
24)
(Polity
IV)2
-0.0
01∗∗∗
-0.0
06∗∗∗
(0.0
00)
(0.0
01)
Contr
ol
vari
able
sayes
yes
yes
yes
yes
yes
#of
countr
ies
156
156
134
134
134
133
133
155
N3,6
78
3,6
78
2,9
29
2,8
06
2,8
04
2,9
08
2,7
41
3,6
55
Pse
udoR
2(M
cFadden
)0.0
01
0.0
06
0.0
60
0.0
57
0.0
71
Notes:
Robust
standard
erro
rsare
dis
pla
yed
inpare
nth
eses
.∗p<
0.1
0,∗∗p<
0.0
5,∗∗∗p<
0.0
1.
aIn
cludes
GD
P/ca
pit
a,
popula
tion
size
,and
gro
wth
rate
.bIn
colu
mn
(4),
EX
EC
,(E
XE
C)2
,natu
ral
reso
urc
ere
nts
,and
dev
elopm
ent
ass
ista
nce
are
inst
rum
ente
dby
thei
rre
spec
tive
lagged
valu
esin
the
pre
vio
us
yea
r.C
olu
mns
(5)
and
(6)
apply
the
sam
elo
gic
for
EX
EC
,(E
XE
C)2
,oil
rents
,dev
elopm
ent
ass
ista
nce
,P
olity
IVand
(Polity
IV)2
.In
all
esti
mati
ons,
Shea
’spart
ial
R2
ranges
bet
wee
n0.5
5and
0.7
8,
leav
ing
litt
leco
nce
rnab
out
pote
nti
ally
wea
kin
stru
men
ts.
22
straints on the executive are measured at a value of 4.8. Further, even perfect democracies are
more likely to suffer from terrorism – a result that is consistent with the theoretical priors and
the empirical results from considering the incidence of terrorism.
Constraints on the executive Development assistance per capita
46
810
% c
hang
e in
like
lihoo
d of
terr
oris
m o
nset
0 2 4 6 8Constraints on executive
05
1015
% c
hang
e in
like
lihoo
d of
terr
oris
m o
nset
0 .02 .04 .06 .08 .1Development assistance in $10,000 per capita
Figure 4: Effect of constraints on the executive and development assistance on the onset ofterrorism, plotting results from column (3) in Table 3.
Related to development assistance per capita, a one standard deviation increase (US$ 117
per capita) relates to a 1.6 percentage point rise in the probability of experiencing terrorism.
In the extreme case, moving from US$ 0 to US$ 1,845 (Jordan in 1979), the onset of terrorism
becomes 25.4 percentage points more likely.
6 Empirical Analysis of Conflicts
After terrorism, I now move to the analysis of domestic conflicts. Recall that the model predicts
a linear negative relationship between constraints on the executive and the incidence and onset
of civil conflict.
6.1 Incidence of Conflicts: Main Results
Table 4 follows the same sequence as analyzing the incidence of terrorism, beginning with a
univariate regression. Indeed, we find a negative link between institutional constraints and the
number of victims from internal conflicts. In terms of magnitude, raising executive constraints
by one level (say, from two to three) is associated with a 12.6 percentage point decrease in
23
the number of conflict casualties. Column (2) shows that this relationship is not quadratic,
in contrast to the relationship with terrorism – a result that is consistent with the theoretical
predictions.
Table 4: OLS regression results, estimating the number of deaths from conflicts in country iand year t.
(1) (2) (3) (4) (5) (6)
Dependent variable: Ln(1+deaths from internal conflict)
EXEC -0.126∗∗ 0.054 -0.145∗∗ -0.108∗ -0.109∗ -0.112∗
(0.061) (0.301) (0.062) (0.058) (0.062) (0.062)
(EXEC)2 -0.021(0.034)
Natural resource rents 0.456 0.507 0.519in US$ 10,000/cap (0.322) (0.363) (0.423)
Country-fixed effects yes yes yes yes
Control variablesa yes yes yes
Year-fixed effects yes yes
Continent-specific time trends yes
# of countries 158 158 158 158 158 158N 3,586 3,586 3,586 3,586 3,586 3,586Adjusted R2 0.015 0.015 0.612 0.614 0.613 0.618
Notes: Standard errors clustered on the country level are displayed in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗
p < 0.01. aIncludes GDP/capita, population size, and growth rate.
Columns (3) to (6) then incorporate country-fixed effects, natural resource rents, the famil-
iar control variables, year-fixed effects, and continent-specific time trends. However, executive
constraints remain a negative predictor of conflict incidence with the respective coefficient only
fluctuating marginally between -0.11 and -0.15. In addition, natural resource rents do not play
any role, consistent with findings from Elbadawi and Sambanis (2002) or Fearon and Laitin
(2003). Remember that according to the profit-maximizing decision by the opposition group
24
modeled in section 3 we should not expect a particularly strong relationship between R and
conflicts, but rather executive constraints should be the dominant factor.
6.2 Incidence of Conflicts: Extensions
As with the analysis of terrorism, I now move to several extensions, displayed in Table 5.
Following the same sequence as in Table 3, columns (1) and (2) consider alternative definitions
of R by incorporating foreign development assistance and oil rents. However, none of these
aspects are closely related to the incidence of national conflict. The negative effect from EXEC,
however, prevails.
Column (3) switches to the polity2 variable as an alternative measure for institutional con-
trols and, as with the analysis of terrorism, the initial result is confirmed. As before, the
more flexible measure of the polity2 variable brings out the underlying relationship with more
statistical precision, as the associated level of statistical significance increases to five percent.
Nevertheless, development assistance and oil rents remain largely irrelevant. In alternative es-
timations, I also address the measurement of the dependent variable. In particular, all results
are preserved when employing a measure for deaths per capita or using a binary indicator for
experiencing 25 or more deaths, which represents a more traditional way of measuring conflict
incidence. These results are referred to the appendix Table AV.18
Finally, columns (4) to (6) re-estimate the corresponding regressions in the familiar IV set-
ting, where executive constraints, natural resource rents, development assistance, oil rents, and
the polity2 variable are instrumented by their lagged values from the previous year. The results
further support the hypotheses, as institutional constraints remain important, but measures for
R do not.
In terms of magnitude, an increase in the level of executive constraints by one point is
associated with a decrease in the number of deaths from conflict by 11 to 13.6 percentage
points, depending on which regression we choose from Table 5. It is also noteworthy to point out
that the corresponding regressions are able to explain approximately 60 percent of the observed
variation in deaths from conflicts throughout the sample, as indicated by the respective adjusted
18Since all regressions are estimated in a fixed-effects framework, I refrain from using a logit or probit approachto estimate the binary outcome variable of conflict incidence, but rather employ a conventional OLS approach,as is common in the literature (see Greene, 2004).
25
Table 5: OLS regression results from extensions, estimating the number of deaths from conflictsin country i and year t.
IV regressionsb
(1) (2) (3) (4) (5) (6)
Dependent variable: Ln(1+deaths from internal conflict)
EXEC -0.115∗ -0.112∗ -0.136∗∗ -0.135∗∗
(0.063) (0.063) (0.068) (0.066)
Natural resource rents 1.264∗ 0.992in US$ 10,000/cap (0.709) (0.794)
Development assistance -2.787 -2.401 -7.307 5.584 5.929 -2.214in US$ 10,000/cap (7.628) (7.585) (8.338) (13.829) (14.257) (14.496)
Oil rents 1.265 1.164 0.966 0.814in US$ 10,000/cap (0.858) (0.827) (0.908) (0.878)
Polity IV -0.050∗∗ -0.056∗∗
(0.023) (0.025)
Control variablesa yes yes yes yes yes yes
Country-fixed effects yes yes yes yes yes yes
Year-fixed effects yes yes yes yes yes yes
Continent-specific time trends yes yes yes
# of countries 136 136 135 135 135 134N 2,880 2,815 2,781 2,847 2,794 2,759Adjusted R2 0.609 0.611 0.593 0.606 0.605 0.586
Notes: Standard errors clustered on the country level are displayed in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗
p < 0.01. aIncludes GDP/capita, population size, and growth rate. bIn column (4), EXEC, natural resource
rents, and development assistance are instrumented by their respective lagged values in the previous year.
Columns (5) and (6) apply the same logic for EXEC, oil rents, development assistance, and Polity IV. In all
estimations, Shea’s partial R2 ranges between 0.27 and 0.78, leaving little concern about potentially weak
instruments.
26
R2 values.
6.3 Onset of Conflicts
Finally, Table 6 turns to the onset of domestic conflicts. For this measure, the statistical variation
throughout the sample diminishes substantially, as conflict has begun in “only” 96 country-year
observations, where the respective country has not suffered from conflict in the preceding year.
In fact, (luckily) only 51 countries appear on this list.
Table 6: Results from probit regressions, estimating the onset of conflict in country i and yeart. Displaying marginal effects.
IV regressionsb
(1) (2) (3) (4) (5) (6) (7)
EXEC -0.010∗∗∗ -0.007∗∗∗ -0.006∗∗∗ -0.066∗∗ -0.060∗∗
(0.002) (0.002) (0.002) (0.029) (0.029)
Natural resource rents 0.027∗ 0.363∗ 0.139in US$ 10,000/cap (0.015) (0.200) (0.165)
Development assistance 0.453 0.507 0.133 4.804 6.012in US$ 10,000/cap (0.654) (0.655) (0.663) (11.158) (11.101)
Oil rents 0.039∗ 0.044∗∗ 0.508∗
in US$ 10,000/cap (0.020) (0.019) (0.259)
Polity IV -0.001∗ -0.012(0.001) (0.008)
Control variablesa yes yes yes yes yes
# of countries 151 129 129 128 127 127 150N 2,918 2,268 2,220 2,205 2,239 2,200 2,873Pseudo R2 (McFadden) 0.046 0.093 0.092 0.088
Notes: Robust standard errors are displayed in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01. aIncludes
GDP/capita, population size, and growth rate. bIn column (5), EXEC, natural resource rents, and development
assistance are instrumented by their respective lagged values in the previous year. Columns (6) and (7) apply the
same logic for EXEC, oil rents, development assistance, Polity IV, and natural resource rents. In all estimations,
Shea’s partial R2 ranges between 0.65 and 0.93, leaving little concern about potentially weak instruments.
Column (1) displays results from a univariate regression, suggesting that the onset of conflict
is less likely for inclusive institutions. This result is confirmed once additional control variables
are included, but the related magnitude decreases to 0.007. Note that natural resource abun-
27
dance is associated with a higher likelihood of conflict onset, something that is not necessarily
suggested by the model, but has been proposed by previous studies (Collier and Hoeffler, 1998,
Koubi et al., 2014).19 Thus, it is possible that further dynamics related to natural resources
relate to civil conflicts.
Further, columns (3) and (4) explore different measures for σ and R. Concerning institutional
constraints, the negative link to conflict is confirmed. Related to available rents, we do confirm
the importance of oil rents in driving up the likelihood of conflict. Again, this indicates that
natural resources could exhibit an additional dynamic when it comes to violent conflicts. In
particular, the results related to natural resources and specifically oil confirm findings by Collier
and Hoeffler (2004), Humphreys (2005), and Ross (2006).
Finally, columns (5) to (7) estimate the familiar sequence of IV regressions. Most impor-
tantly, executive controls remain a powerful predictor of conflict onset and the associated co-
efficient increases more than ten-fold, from -0.006 to -0.066. This indicates that applying a
regular probit framework would underestimate the effect of executive constraints on the onset
of conflict. Intuitively, it is possible that an outbreak of open conflict leads to a tightening of
institutional controls, as the ruling government tries to maintain control. Such dynamics would
make it difficult to isolate the true effect of institutional controls on the onset of conflict in a
standard OLS framework. However, employing executive constraints in the preceding (peaceful)
year as an instrument for contemporaneous executive controls circumvents this problem and is
likely better able to reveal the underlying effect of executive constraints on conflict onset.
In addition, increased oil rents continue to predict conflict onset, whereas foreign development
assistance remains an irrelevant factor. Finally, employing the Polity2 variable as an alternative
estimate for institutional controls confirms the negative impact on conflict onset, but the derived
coefficient fails to clear the conventional levels of statistical significance. It is important to note
that the coefficient becomes stronger in quantitative terms, but standard errors are inflated
substantially (from 0.001 to 0.008). In fact, we observe substantially elevated standard errors
for all derived coefficients in the IV regressions – a result that is likely driven by less statistical
variation in the employed instruments and the limited number of observations in which conflict
19Nillesen and Bulte (2014) provide a recent summary of the link between natural resources and violent conflict.Also, van der Ploeg (2011) provides an overview of how natural resources can affect development in various ways.
28
emerges (96).
7 Conclusion
This paper analyzes the decision of an opposition group to pursue a campaign of organized
violence against a ruling government. Depending on constraints on the executive (i.e., the
inclusiveness of political institutions) and the availability of rents, the group can choose between
peace, terrorism, and open civil conflict. An important distinction between open conflict and
terrorism comes from comparing the associated costs and benefits. Contrary to civil conflict,
both parameters are naturally limited for a terrorist campaign.
Analyzing the opposition group’s optimal choice then reveals that terrorism becomes likely
if executive constraints are intermediate and rents are high. In fact, even in largely inclusive
institutions terrorism remains a viable option, but only if rents are considerable. Civil conflict
emerges as the dominant option if executive constraints are particularly poor, whereas peace
becomes the likely outcome under high executive constraints and a modest to low availability of
rents.
Taking these hypotheses to the data, the paper analyzes 5,400 and 3,586 country-year ob-
servations for terrorism and conflicts, respectively. Employing country- and year-fixed effects,
continent-specific time trends, and the conventional time-variant control variables, the theo-
retical predictions receive firm support. Intermediate ranges of executive control increase the
number of deaths from terrorism and the likelihood of terrorism onset. However, less constraints
on the executive likely pushes the country into civil conflict.
Further, large available rents increase the incidence of terrorism, as well as the likelihood of
terrorism onset. For all three rent measures (natural resource rents, oil rents, and development
assistance), a one standard deviation increase in per capita revenue is suggested to increase the
number of deaths from terrorism by approximately 15 to 16 percentage points. It is remark-
able how consistent this magnitude remains for all three measures and across several different
estimations.
Related to domestic conflicts, tighter controls on the executive substantially decrease the
number of casualties in a linear fashion. Moving from a totally authoritarian regime to perfectly
29
inclusive institutions is associated with a decrease in the number of battle-related deaths by
approximately 74 percentage points. Considering conflict onset, results are less precise in statis-
tical terms – an artifact that becomes less surprising once we are reminded of the rare occurrence
of conflict onset (96 observations in 51 countries). Nevertheless, the corresponding coefficients
consistently confirm the notion that executive constraints are negatively tied to conflict onset
in a linear way.
To my knowledge, this paper is among the first to jointly analyze the profit-maximizing
decision of an opposition group between peace, terrorism, and open civil conflict. The theoretical
motivation is basic and one could think of several extensions. Nevertheless, the empirical part
of the paper shows that the model’s simple predictions are consistently observed in global data,
even when controlling for a number of potentially confounding factors and fixed effects. As
such, the paper may serve as a starting point to better understand and anticipate the decisions
of opposition groups, integrating the drivers of terrorism and national conflicts.
30
References
Abadie, A. (2006). Poverty, political freedom, and the roots of terrorism. American EconomicReview Papers & Proceedings, 96(2):50–56.
Acemoglu, D., Johnson, S., and Robinson, J. A. (2005). Institutions as a fundamental cause oflong-run growth. Handbook of Economic Growth, 1:385–472.
Acemoglu, D., Johnson, S., Robinson, J. A., and Yared, P. (2008). Income and democracy.American Economic Review, 98(3):808–842.
Acemoglu, D., Vindigni, A., and Ticchi, D. (2010). Persistence of civil wars. Journal of theEuropean Economic Association, 8(2-3):664–676.
Acemoglu, D. and Wolitzky, A. (2014). Cycles of conflict: An economic model. The AmericanEconomic Review, 104(4):1350–1367.
Arezki, R. and Bruckner, M. (2011). Oil rents, corruption, and state stability: Evidence frompanel data regressions. European Economic Review, 55(7):955–963.
Besley, T. and Persson, T. (2009). Repression or civil war? American Economic Review Papers& Proceedings, 99(2):292–297.
Besley, T. and Persson, T. (2011). The logic of political violence. The Quarterly Journal ofEconomics, 126(3):1411–1445.
Bhattacharyya, S. and Hodler, R. (2010). Natural resources, democracy and corruption. Euro-pean Economic Review, 54(4):608–621.
Blattman, C. and Miguel, E. (2010). Civil war. Journal of Economic Literature, 48(1):3–57.
Blomberg, S. B., Hess, G. D., and Weerapana, A. (2004). Economic conditions and terrorism.European Journal of Political Economy, 20(2):463–478.
Cervellati, M., Jung, F., Sunde, U., and Vischer, T. (2014). Income and democracy: Comment.The American Economic Review, 104(2):707–719.
Chenoweth, E. (2013). Terrorism and democracy. Annual Review of Political Science, 16:355–378.
CIA (2016). CIA World Factbook. The World Factbook. Retrieved fromhttps://www.cia.gov/library/publications/the-world-factbook/.
Clutterbuck, R. (1992). Negotiating with terrorists. Terrorism and Political Violence, 4(4):263–287.
Collier, P. and Hoeffler, A. (1998). On economic causes of civil war. Oxford Economic Papers,50(4):563–573.
Collier, P. and Hoeffler, A. (2004). Greed and grievance in civil war. Oxford Economic Papers,56(4):563–595.
31
Collier, P., Hoeffler, A., and Rohner, D. (2009). Beyond greed and grievance: Feasibility andcivil war. Oxford Economic Papers, 61(1):1–27.
Cotet, A. M. and Tsui, K. K. (2013). Oil and conflict: What does the cross country evidencereally show? American Economic Journal: Macroeconomics, 5(1):49–80.
Dixon, J. (2009). What causes civil wars? Integrating quantitative research findings. Interna-tional Studies Review, 11(4):707–735.
Do, Q.-T. and Iyer, L. (2010). Geography, poverty and conflict in Nepal. Journal of PeaceResearch, 47(6):735–748.
Dolnik, A. and Fitzgerald, K. M. (2011). Negotiating hostage crises with the new terrorists.Studies in Conflict & Terrorism, 34(4):267–294.
Elbadawi, I. and Sambanis, N. (2002). How much war will we see? Explaining the prevalenceof civil war. Journal of Conflict Resolution, 46(3):307–334.
Enders, W. and Hoover, G. A. (2012). The nonlinear relationship between terrorism and poverty.American Economic Review Papers & Proceedings, 102(3):267–272.
Esteban, J., Mayoral, L., and Ray, D. (2012). Ethnicity and conflict: An empirical study.American Economic Review, 102(4):1310–42.
Esteban, J. and Ray, D. (2008). On the salience of ethnic conflict. American Economic Review,98(5):2185–2202.
Esteban, J. and Ray, D. (2011). Linking conflict to inequality and polarization. The AmericanEconomic Review, 101(4):1345–1374.
Fearon, J. D. and Laitin, D. D. (2003). Ethnicity, insurgency, and civil war. American PoliticalScience Review, 97(01):75–90.
Findley, M. G. and Young, J. K. (2012). Terrorism and civil war: A spatial and temporalapproach to a conceptual problem. Perspectives on Politics, 10(02):285–305.
Greene, W. (2004). The behaviour of the maximum likelihood estimator of limited dependentvariable models in the presence of fixed effects. The Econometrics Journal, 7(1):98–119.
Group, W. B. (2012). World Development Indicators 2012. World Bank Publications.
Hegre, H. (2014). Democracy and armed conflict. Journal of Peace Research, 51(2):159–172.
Hegre, H. and Sambanis, N. (2006). Sensitivity analysis of empirical results on civil war onset.Journal of Conflict Resolution, 50(4):508–535.
Humphreys, M. (2005). Natural resources, conflict, and conflict resolution: Uncovering themechanisms. Journal of Conflict Resolution, 49(4):508–537.
Islam, N. (1995). Growth empirics: A panel data approach. The Quarterly Journal of Economics,110(4):1127–1170.
32
Koubi, V., Spilker, G., Bohmelt, T., and Bernauer, T. (2014). Do natural resources matter forinterstate and intrastate armed conflict? Journal of Peace Research, 51(2):227–243.
Krueger, A. B. and Maleckova, J. (2003). Education, poverty and terrorism: Is there a causalconnection? Journal of Economic Perspectives, 17(4):119–144.
LaFree, G. and Dugan, L. (2007). Introducing the Global Terrorism Database. Terrorism andPolitical Violence, 19(2):181–204.
Lessing, B. (2015). Logics of violence in criminal war. Journal of Conflict Resolution, 59(8):1486–1516.
Marshall, M. G. and Jaggers, K. (2002). Polity IV project: Political regime characteristics andtransitions, 1800-2002.
Miguel, E., Satyanath, S., and Sergenti, E. (2004). Economic shocks and civil conflict: Aninstrumental variables approach. Journal of Political Economy, 112(4):725–753.
Mirestean, A. and Tsangarides, C. G. (2016). Growth determinants revisited using Limited-Information Bayesian Model Averaging. Journal of Applied Econometrics, 31(1):106–132.
Neumann, P. R. (2007). Negotiating with terrorists. Foreign Affairs – New York, 86(1):128.
Nielsen, R. A., Findley, M. G., Davis, Z. S., Candland, T., and Nielson, D. L. (2011). Foreign aidshocks as a cause of violent armed conflict. American Journal of Political Science, 55(2):219–232.
Nillesen, E. and Bulte, E. (2014). Natural resources and violent conflict. Annu. Rev. Resour.Econ., 6(1):69–83.
Nunn, N. and Qian, N. (2014). US food aid and civil conflict. The American Economic Review,104(6):1630–1666.
Reynal-Querol, M. and Montalvo, J. G. (2005). Ethnic polarization, potential conflict and civilwar. American Economic Review, 95(3):796–816.
Ross, M. (2006). A closer look at oil, diamonds, and civil war. Annu. Rev. Polit. Sci., 9:265–300.
Schularick, M. and Steger, T. M. (2010). Financial integration, investment, and economic growth:Evidence from two eras of financial globalization. The Review of Economics and Statistics,92(4):756–768.
Schutte, S. (2015). Geography, outcome, and casualties: A unified model of insurgency. Journalof Conflict Resolution, 59(6):1101–1128.
Shea, J. (1997). Instrument relevance in multivariate linear models: A simple measure. Reviewof Economics and Statistics, 79(2):348–352.
START (2015). Global terrorism database. National Consortium for the Study of Terrorismand Responses to Terrorism (START). Retrieved from http://www.start.umd.edu/gtd.
33
Temple, J. (1999). The new growth evidence. Journal of Economic Literature, 37(1):112–156.
UCDP (2015). UCDP battle-related deaths dataset v.5-2015, Uppsala Conflict Data Program.www.ucdp.uu.se, Uppsala University.
van der Ploeg, F. (2011). Natural resources: Curse or blessing? Journal of Economic Literature,49(2):366–420.
Weidmann, N. B. and Ward, M. D. (2010). Predicting conflict in space and time. Journal ofConflict Resolution, 54(6):883–901.
34
Table AI: Payoff structure, modelling the incumbent’s and opposition’s options.
Incumbent
σL σH
Opposition
Civil conflict βR(1− σL
), βR
(1− σH
),(
1− σL)(
1− β)R
(1− σH
)(1− β
)R
Terrorism σLR+ β(12 − σL)R− c, σHR+ β(1
2 − σH)R− c,(1− σL)R− β(1
2 − σL)R (1− σH)R− β(12 − σH)R
Peace σLR, σHR,
(1− σL)R (1− σH)R
Appendix
The Government’s Choice
It is straightforward to show that the government always chooses the lowest value of σ possible,
given constitutional constraints. To see this, suppose the incumbent group can choose between
σL, the minimal share dedicated to the opposition as per institutional constraints, and σH (with
0 ≤ σL < σH). In that case, Table AI displays the corresponding payoff matrix.
From the incumbent’s perspective, σL constitutes the dominant strategy. In the case of
war, the incumbent’s income is higher for σL, since σL < σH . In the case of terrorism, the
corresponding payoff is larger for σL. Specifically, for terrorism: (1 − σL)R − β(12 − σL)R >
(1 − σH)R − β(12 − σH)R, which, after collecting terms, produces σH > σL. Finally, in the
peaceful scenario, the incumbent’s income from setting σL also dominates, as comparing the
respective payoffs again produces σH > σL.
35
Sam
ple
Cou
ntr
ies
Tab
leA
II:
Cou
ntr
ies
incl
ud
edin
terr
oris
man
dco
nfl
ict
sam
ple
wit
hre
spec
tive
nu
mb
erof
annu
alob
serv
atio
ns.
Countr
yT
err
ori
smC
onfl
ict
Countr
yT
err
ori
smC
onfl
ict
Countr
yT
err
ori
smC
onfl
ict
Countr
yT
err
ori
smC
onfl
ict
Sam
ple
Sam
ple
Sam
ple
Sam
ple
Sam
ple
Sam
ple
Sam
ple
Sam
ple
Afg
hanis
tan
11
11
Dom
inic
an
Republic
43
25
Latv
ia18
18
Rw
anda
43
25
Alb
ania
29
25
Ecuador
43
25
Lebanon
24
25
Saudi
Ara
bia
42
25
Alg
eri
a43
25
Egypt,
Ara
bR
ep.
43
25
Leso
tho
43
25
Senegal
43
25
Arg
enti
na
43
25
El
Salv
ador
43
25
Lib
eri
a43
25
Serb
ia8
8A
rmenia
17
17
Equato
rial
Guin
ea
22
15
Lib
ya
14
14
Sie
rra
Leone
43
25
Aust
ralia
43
25
Eri
trea
18
19
Lit
huania
10
10
Slo
vak
Republi
c20
21
Aust
ria
43
25
Est
onia
18
18
Luxem
bourg
14
14
Slo
venia
18
18
Azerb
aij
an
16
16
Eth
iopia
31
25
Macedonia
,F
YR
20
21
Solo
mon
Isla
nds
22
23
Bahra
in32
25
Fij
i43
25
Madagasc
ar
43
25
South
Afr
ica
43
25
Bangla
desh
41
25
Fin
land
43
25
Mala
wi
43
25
Spain
43
25
Bela
rus
21
22
Fra
nce
43
25
Mala
ysi
a43
25
Sri
Lanka
43
25
Belg
ium
14
14
Gab
on
43
25
Mali
43
25
Sudan
33
Benin
43
25
Gam
bia
,T
he
43
25
Mauri
tania
43
25
Suri
nam
e37
25
Bhuta
n32
25
Georg
ia16
16
Mauri
tius
36
25
Sw
aziland
42
25
Bolivia
43
25
Germ
any
23
24
Mexic
o43
25
Sw
eden
43
25
Bots
wana
43
25
Ghana
43
25
Mold
ova
19
19
Sw
itzerl
and
32
25
Bra
zil
43
25
Gre
ece
43
25
Mongolia
31
25
Syri
an
Ara
bR
epublic
37
19
Bulg
ari
a32
25
Guate
mala
43
25
Monte
negro
88
Taji
kis
tan
16
16
Burk
ina
Faso
43
25
Guin
ea
26
25
Moro
cco
43
25
Tanzania
24
25
Buru
ndi
43
25
Guin
ea-B
issa
u39
25
Mozam
biq
ue
32
25
Thailand
43
25
Cab
oV
erd
e32
25
Guyana
43
25
Nam
ibia
23
24
Tim
or-
Lest
e2
2C
am
bodia
20
20
Hait
i15
15
Nepal
43
25
Togo
43
25
Cam
ero
on
43
25
Hondura
s43
25
Neth
erl
ands
43
25
Tri
nid
ad
and
Tobago
43
25
Canada
43
25
Hungary
21
22
New
Zeala
nd
35
25
Tunis
ia43
25
Centr
al
Afr
ican
Republic
43
25
India
43
25
Nic
ara
gua
43
25
Turk
ey
43
25
Chad
43
25
Indonesi
a43
25
Nig
er
43
25
Turk
menis
tan
77
Chile
43
25
Iran,
Isla
mic
Rep.
41
23
Nig
eri
a43
25
Uganda
30
25
Chin
a43
25
Iraq
30
12
Norw
ay
43
25
Ukra
ine
19
19
Colo
mbia
43
25
Irela
nd
42
25
Om
an
43
25
Unit
ed
Ara
bE
mir
ate
s37
25
Com
oro
s32
25
Isra
el
43
25
Pakis
tan
43
25
Unit
ed
Kin
gdom
43
25
Congo,
Dem
.R
ep.
41
22
Italy
43
25
Panam
a43
25
Unit
ed
Sta
tes
43
25
Congo,
Rep.
43
25
Jam
aic
a43
25
Papua
New
Guin
ea
38
25
Uru
guay
43
25
Cost
aR
ica
43
25
Japan
43
25
Para
guay
43
25
Uzb
ekis
tan
15
15
Cote
d’I
voir
e43
25
Jord
an
37
25
Peru
43
25
Venezuela
,R
B43
25
Cro
ati
a18
18
Kazakhst
an
21
22
Philip
pin
es
43
25
Vie
tnam
28
25
Cuba
42
25
Kenya
43
25
Pola
nd
22
23
Yem
en,
Rep.
20
21
Cypru
s37
25
Kore
a,
Rep.
43
25
Port
ugal
43
25
Zam
bia
43
25
Czech
Republic
20
21
Kuw
ait
18
18
Qata
r13
13
Zim
babw
e43
25
Denm
ark
43
25
Kyrg
yz
Republic
17
17
Rom
ania
22
23
Dji
bouti
22
23
Lao
PD
R28
25
Russ
ian
Federa
tion
21
22
36
Summary Statistics
Table AIII: Summary statistics for terrorism sample (1970 – 2014, excluding 1993 because ofdata unavailability). 5,400 observations unless indicated otherwise.
Variable Mean Min. Sourcea Description(Std. Dev.) (Max.)
Deaths from terrorism 43.79 0 GTD Number of deaths from terrorist attacks in country i and(261.29) (7,038) year t; applying ln(1+deaths)
EXEC 4.38 1 Polity IV Variable EXCONST , executive constraints,(2.31) (7) ranging from 1 to 7
Natural resource rents 0.06 0 WDI Total natural resource rents in US$ 10,000 per capitain US$ 10,000/cap (0.24) (4.63) (adjusted by GDP in constant 2005 prices and
population size); initially natural resource rents in % of GDP
GDP/capita 7,823 70 WDI GDP per capita in constant 2005 US$;(12,750) (87,773) applying Ln(GDP/capita)
Population size 3,909 23.29 WDI Total population size; applying Ln(GDP/capita)in 10,000 (13,370) (135,738)
Growth rate 2.03 -62.21 WDI GDP per capita growth(5.76) (104.66)
Polity IV 11.88 0 Polity IV Variable POLITY 2, re-scaled to run between 0(7.28) (20) (total autocracy) and 20 (total democracy); 5,366 observations
Development assis- 0.01 0 WDI Net official development assistance and official aidtance in US$10,000/cap
(0.01) (0.18) received (constant US$2005) in US$ 10,000 per capita
(adjusted by population size); 4,333 observations
Oil rents in 0.04 0 WDI Oil rents in US$ 10,000 per capita (adjusted by GDP inUS$ 10,000/cap (0.19) (4.53) constant 2005 prices and population size); initially oil
rents in % of GDP; 4,190 observations
Notes: aData come from the Global Terrorism Database (GTD), Polity IV, and the World Development Indicators
provided by the World Bank (WDI).
37
Table AIV: Summary statistics for conflict sample (1989 – 2014). 3,586 observations unlessindicated otherwise.
Variable Mean Min. Sourcea Description(Std. Dev.) (Max.)
Deaths from internal 149.13 0 UCDP Number of deaths from internal and internationalizedconflict (1,123) (49,698) internal conflicts in country i and year t; applying
ln(1+deaths)
EXEC 4.84 1 Polity IV Variable EXCONST , executive constraints,(2.09) (7) ranging from 1 to 7
Natural resource rents 0.07 0 WDI Total natural resource rents in US$ 10,000 per capitain US$ 10,000/cap (0.24) (3.29) (adjusted by GDP in constant 2005 prices and
population size); initially natural resource rents in% of GDP
GDP/capita 8,710 70 WDI GDP per capita in constant 2005 US$;(13,911) (87,773) applying Ln(GDP/capita)
Population size 4,102 32 WDI Total population size; applying Ln(population size)in 10,000 (14,107) (135,738)
Growth rate 2.18 -62.21 WDI GDP per capita growth(5.81) (104.66)
Polity IV 13.44 0 Polity IV Variable POLITY 2, re-scaled to run between 0(6.54) (20) (total autocracy) and 20 (total democracy); 3,551 observations
Development assis- 0.01 0 WDI Net official development assistance and official aidtance in US$10,000/cap
(0.01) (0.09) received (constant US$2005) in US$ 10,000 per capita
(adjusted by population size); 2,880 observations
Oil rents in 0.04 0 WDI Oil rents in US$ 10,000 per capita (adjusted by GDP inUS$ 10,000/cap (0.14) (1.91) constant 2005 prices and population size); initially oil
rents in % of GDP; 2,815 observations
Notes: aData come from the UCDP Battle-Related Deaths Dataset version 5.0-2015 (available under
http://www.pcr.uu.se/research/ucdp/datasets/ucdp_battle-related_deaths_dataset/), Polity IV, and the World
Development Indicators provided by the World Bank (WDI).
38
Tab
leA
V:
OL
Sre
gres
sion
resu
lts
from
exte
nsi
ons,
esti
mat
ing
the
nu
mb
erof
dea
ths
from
terr
oris
man
dd
omes
tic
con
flic
tin
cou
ntr
yi
and
yea
rt.
Col
um
ns
(1)
thro
ugh
(4)
use
dea
ths
per
cap
ita
and
colu
mn
s(5
)th
rou
gh(8
)u
sea
bin
ary
ind
icato
rfo
r25
orm
ore
dea
ths
asth
ed
epen
den
tva
riab
le.
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
Dep
enden
tva
riable
:Ln( 1+d
eath
sfrom
terroris
mpopula
tion
)L
n( 1+d
eath
sfrom
dom
esticconflict
popula
tion
)25+
terr
ori
smdea
ths
(0/1)
25+
conflic
tdea
ths
(0/1)
EX
EC
0.3
18∗
0.2
98∗
-0.1
15∗
-0.1
12∗
0.0
61∗
0.0
59∗
-0.0
21∗∗
-0.0
21∗∗
(0.1
72)
(0.1
75)
(0.0
63)
(0.0
63)
(0.0
32)
(0.0
32)
(0.0
11)
(0.0
11)
(EX
EC
)2-0
.040∗
-0.0
38∗
-0.0
08∗∗
-0.0
08∗∗
(0.0
21)
(0.0
22)
(0.0
04)
(0.0
04)
Natu
ral
reso
urc
ere
nts
0.7
61∗∗∗
1.2
64∗
0.0
78∗
0.1
39
inU
S$
10,0
00/ca
p(0
.279)
(0.7
09)
(0.0
45)
(0.1
11)
Dev
elopm
ent
ass
ista
nce
12.7
01∗∗
12.7
29∗∗
-2.7
87
-2.4
01
1.9
77∗∗
1.9
89∗∗
-0.4
55
-0.3
93
inU
S$
10,0
00/ca
p(6
.082)
(6.1
46)
(7.6
28)
(7.5
85)
(0.8
51)
(0.8
62)
(1.3
77)
(1.3
71)
Oil
rents
0.7
61∗∗
1.2
65
0.0
76
0.1
18
inU
S$
10,0
00/ca
p(0
.299)
(0.8
58)
(0.0
48)
(0.1
34)
Contr
ol
vari
able
sayes
yes
yes
yes
yes
yes
yes
yes
Countr
y-fi
xed
effec
tsyes
yes
yes
yes
yes
yes
yes
yes
Yea
r-fixed
effec
tsyes
yes
yes
yes
yes
yes
yes
yes
Conti
nen
t-sp
ecifi
cyes
yes
yes
yes
yes
yes
yes
yes
tim
etr
ends
#of
countr
ies
136
136
136
136
136
136
136
136
N4,3
33
4,1
90
2,8
80
2,8
15
4,3
33
4,1
90
2,8
80
2,8
15
Adju
sted
R2
0.5
51
0.5
55
0.6
16
0.6
18
0.4
13
0.4
19
0.5
59
0.5
58
Notes:
Sta
ndard
erro
rscl
ust
ered
on
the
countr
yle
vel
are
dis
pla
yed
inpare
nth
eses
.∗p<
0.1
0,∗∗p<
0.0
5,∗∗∗p<
0.0
1.
aIn
cludes
GD
P/ca
pit
a,
popula
tion
size
,and
gro
wth
rate
.
39